Еволюція просторово-часових методів знешумлення цифрового відео

Abstract

The work reviews the evolution of spatio-temporal video denoising methods, from classical nonlocal approaches to modern neural network-based models. Particular attention is given to V-BM3D and V-BM4D models, including the mathematical description of the noisy video model, collaborative filtering, and weighted aggregation of estimates. The work further highlights the transition to deep learning methods that use temporal dependencies between neighboring frames for more effective video restoration